Advanced Structural Equation Modelling

Day 1: We start by embedding exploratory factor analysis into SEModels. To date, SEM facilities allowed for the testing of confirmatory models. New developments in the field incorporated the exploratory logic into structural models allowing for more precise definition of factors and improved reliability and validity assessments of concepts that are not sufficiently established theoretically.  In the second session we will look at how SEM can, in addition to finding latent variable, reduce systematic measurement bias associated with the methods selected for data collection such as filtering out extreme response bias, acquiescence bias, and etc.

Day 2: This class we do an overview of the flexibility that modern SEM developments gave users including the incorporation of various different distributions, estimators, even sampling weights and clustered standard errors.  We especially focus on the relationship between Item Response Models (see also Psychometrics class) and Confirmatory Factor Analyses with logit and probit link functions.  We also explore how, in addition to observed heterogeneity through multiple groups, SEM now can incorporate latent heterogeneity into the framework.

Day 3: First session we explore how SEM deals with missing data. We develop a theoretically founded understanding of how to think about missing data and how we can alleviate bias associated with missing data in SEModels.  The second session will focus on specifying models with interactions between latent variables.  While specifying interactions between observed variables is relatively simple in the regression framework (which extends directly to SEM), interactions between latents is not so straightforward.

Day 4: We get to longitudinal models first exploring AR1 structures, extending into models where random measurement error is flushed from the observations over time. Through this approach it is possible to specify truly causality models under less restrictive assumptions than traditionally in the SEM framework. Then we continue with models of change where change, its nonlinear trends and their predictors can be simultaneously modelled with what is most often referred to as the latent growth curve model.

Day 5: Briefly introduces how the logic of growth curve models are analogous with multilevel growth models and point to a framework (notably different, though equivalent, from what is covered in the multilevel SEM class) that allows for the modeling of data on multiple levels of analysis.  Finally we close by the introduction of additional model constraints to model specification and demonstrate their power in simplifying or extending model specifications through multiple examples.

It is very important that any participant of this class has a solid understanding of the basics of Structural Equation Modeling at least to the level of the Winter School course in SEM.  You need to know path models, confirmatory factor models and full structural models, know how to build and evaluate structural equation models and assess their fit and results.  In absence of this knowledge I strongly advise you to take the Introductory Structural Equation Modeling class by the Cora Maas Award winning professor Jochen Mayerl.

Additionally, some of the models we will discuss are only available in Mplus.  For this reason we will use Mplus for the purposes of the class (though for several topics presented other software, like the free and open source Lavaan in R, is also usable) basic knowledge of Mplus is also very important.  If you are not familiar with Mplus, I offer an Mplus software class in the pre-session.  

This course is placed in sequence with the Multiulevel Structural Equation Modeling class during the first week.  While it is not a must, I would recommend you also check that out along with the Mplus class.